Contrapunctus XXV: the DKT 3-node triangular element for composite

plates and shells

Preface

As Prof. Reddy states in the preface of his “Mechanics of laminated composite plates and
shells” [1], there does not exist another book that contains a detailed coverage of various
laminate theories, analytical solutions and finite element models (seems that this statement
continues true). Prof. Reddy himself gives very little attention to triangular elements, making
a brief mention of a simple 9-dof bending triangular element, developed by Bazeley,
Cheung, Irons, and Zienkiewicz, the BICZ element. Incidentally, the BCIZ element would have
an important drawback due to its incompatibility: it would not converge to the exact
solution for some mesh patterns.

There is therefore an important window of opportunity to propose, for composite plates and
shells, a simple triangular element, taking into consideration the three translations and
three rotations at the triangle vertices. Such an element would be easy to understand and
simple to implement on a micro- or mini-computer, and always give a stable and correct
solution. This paper introduces a lean and efficient triangular element constructed on the 9-
dof bending DKT theory. The resulting 15-dof plate element is extremely accurate and
converges well to the analytical results given by the Navier solution of rectangular angle-ply
laminated plates. A geometrical matrix consistent with the DKT bending element is also
proposed, and its excellent performance is illustrated. Finally, the 18-dof DKT shell element
compares well with the TRIC element of Teneketzis and Argyris [2] in the three famous
benchmarks: the Scordelis-Lo roof, the pinched cylinder, and the hemispherical shell with
point loads.

Scholars often times rate flat shell triangular elements as crude, which is perhaps not fair,
since they offer extremely attractive and versatile opportunities for practicing engineers in
defining irregular shaped geometries using small computers. The DKT triangular shell
element for composites is easy to implement on such small computers. All the numerical
integrations can be easily accomplished with just three integration points located at the
triangle mid-nodes. Also, with only 18-dof per element, the size of the global stiffness matrix
equation to be solved is reduced to a minimum, and, with the Cholesky algorithm applicable,
the macro run time reduction is fivefold.

We do not know of any other element with such attractiveness to engineers: broadly
applicable, simple to implement, secure in its results, and with a fast response on a small
computer, inclusive on Excel enhanced with macros to assemble and solve the global matrix
equation.

Edouard Zurstrassen
Rio Claro (SP), Brazil
[email protected]

1
Abstract

This paper presents lean and efficient triangular elements constructed on the 9-dof bending
DKT theory enhanced to composite materials. The resulting 15-dof plate element is
extremely accurate and converges well to the analytical results given by the Navier solution
of rectangular angle-ply laminated plates. Introducing the drilling degrees of freedom, an 18-
dof composite shell element is proposed. Various numerical results obtained with those
elements are compared with examples found in Reddy’s book [1].

The paper is divided in four sections:

A. The DKT Bending Element (3-node, 3-dof per node element): tutorial structured in
well-defined parts with an easily understandable sequence of topics, it shall
facilitate the reader to implement the element on its mini- or micro-computer. The
equations consider composite materials.
B. Coupling in composite plates with the DKT bending element (3-node, 5-dof per node
element): this section considers plates submitted to bending along with membrane
deformation. The coupling mechanism between bending and membrane
deformation, peculiar to composites, is taken into account with the development of
two coupling stiffness matrices.
C. Buckling of composite plates with a consistent geometrical stiffness matrix (3-node,
5-dof per node element).
D. Composite shells FEA with the DKT element (3-node, 6-dof per node element):
adding drilling stiffnesses to the previous element, an easily implementable element
for composite shells is proposed.

2
Introduction

It is notably difficult to design a simple triangular element, which allows analyzing shells of
relative complexity without resulting in enormous global matrix equations to be solved, and
that yields always a stable and correct solution. We allude to the ideal triangular element
with only the 18 essential degrees of freedom, which are the three translations and the
three rotations at the triangle vertices. Such simplicity is required for an efficient and
economical implementation on mini- or micro-computers of finite element models, which
enable the analysis of isotropic and laminated composite shells of arbitrary geometry. The
difficulty to develop such simple triangular element can be illustrated by the book "Finite
Element Analysis for Composite Structures" of Teneketzis and Argyris [2], who needed a full
book to explain their excellent, but complex, TRIC element. Although the TRIC element
attends the simplicity of 18-dof, it is neither simple to understand nor easy to implement on
a computer. In fact, the foremost challenge to elude the usual deficiencies of simple
triangular elements resides in having in hands a robust bending element as the centerpiece
of the shell element to be built. The Discrete Kirchhoff Triangle (DKT) is probably the best
candidate to fulfill this role. Since the well-known paper of Batoz et al. (1980) [3], the DKT
element has received more and more attention and positive appraisals from the specialized
literature. For instance, the discrete Kirchhoff approach is adopted in the development of all
thin shell elements of the commercial finite element package ABAQUS. In particular, the DKT
element is used to formulate the flat shell element STRI3.

In the discrete Kirchhoff approach, the superior performance of the element is accomplished
by choosing for the rotations, dw/dx and dw/dy, higher interpolation orders than the order
allowed by the desired number of nodes. Approximation for w within the element is neither
stated nor required. Thanks to judicious assumptions, the reduction to the desired number
of nodes is then achieved engineering new shape functions from the original ones. These
assumptions include the imposition of zero transverse shear strains at a discrete number of
points, normally chosen at the nodes considered by the original shape functions.

The DKT bending element presented herein is a 3-node, 9-dof, triangular element. The
resulting element stiffness matrix is therefore 9 x 9 in size, and can be merged with a plane
elasticity 6 x 6 membrane matrix. Furthermore, with the consideration of drilling stiffness,
an effective shell element with a lean 18 x 18 stiffness matrix can be obtained to study thin
composite structures.

[1] J.N. Reddy, "Mechanics of Laminated Composite Plates and Shells – Theory and Analysis",
second edition. CRC Press (2004)

[2] “Finite Element Analysis for Composite Structures” by Lazarus Teneketzis Tenek and John
Argyris. Solid Mechanics and its Application, Volume 59. Kluwer Academic Publishers (1998).

[3] J.L. Batoz, K. J. Bathe and L. W. Ho, “A study of three-node triangular plate bending
element”, Int. j . numer. methods eng., 15, 1771-1812 (1980).

3
A. The DKMT Bending Element

Introduction
The development of a DKT element for composites has pursued the following main
objectives:

a) The bending triangular element is a lean and efficient element with only 9 dof; that
is, the displacement w and the rotations x, and y, at the three corner nodes. It
should address the deficiencies, mainly sensitivity to the mesh design and
singularity, of the 9-dof element referred to as the Kirchhoff triangular plate
element, which we presented in Contrapunctus X.
b) The element should be easily implemented on a computer, and, particularly, on
Excel enhanced with macros to assemble and solve the global matrix equation.
c) The bending element can be used to build an efficient 18-dof triangular flat element
for composite plates and shells.

The bending element would however comply with the following constraints:

a) Since we undertake a Kirchhoff solution for thin plates, the nodal point rotations,
about the x- and y-axes, should be taken as x = dw/dy and y = −dw/dx,
respectively.
b) Since the element models thin plates, as governed by the classical plate theory, the
element stiffness matrix would be derived from the classical plate theory differential
equation, and the Kirchhoff assumptions would be imposed at any discrete points.

The Bending Stiffness Matrix

The first assumption to formulate the DKT bending element is that the rotations about the x-
and y-axes, y = −x, and x = y, respectively, vary quadratically over the element, i.e.

x =  i = 1 to 6Ni.xi y =  i = 1 to 6Ni.yi (1)

where, imposing the Kirchhoff constraints, xi = −(dw/dx)i and yi = −(dw/dy)i are the nodal
values at the vertices and at the mid-nodes. The shape functions in area coordinates,
Ni( ), are well-known for this case, and are given below.

N1 = (1 −  − )(1 − 2 − 2) N4 = 4

N2 = (2 − 1) N5 = 4(1 −  − ) (2)

N3 = (2 − 1) N6 = 4(1 −  − )

Making use of judicious assumptions described in appendix 1, Batoz et al. ("A study of three-
node triangular plate bending elements" - 1980) arrives to formulate the rotations, x and y,
in function of the 9 corner node displacements collected in vector {ue} and new shape
functions Hx and Hy:

x ≈ HTx.{ue} y ≈ HTy.{ue} (3)

{ue} = [w1, x1, y1, w2, x2, y2, w3, x3, y3]T

4
with y = x and x = −y, and

3(a6N6 − a5N5)/2 3(d6N6 − d5N5)/2

b5N5 + b6N6 −N1 + e5N5 + e6N6
N1 − c5N5 − c6N6 − b5N5 − b6N6
3(a4N4 − a6N6)/2 3(d4N4 − d6N6)/2
Hx = b4N4 + b6N6 Hy = −N2 + e4N4 + e6N6 (4)
N2 − c4N4 − c6N6 − b4N4 − b6N6
3(a5N5 − a4N4)/2 3(d5N5 − d4N4)/2
b4N4 + b5N5 −N3 + e4N4 + e5N5
N3 − c4N4 − c5N5 − b4N4 − b5N5

We have thus in hands a total of 9 pairs of new shape functions to approximate, in functions
of the 9 nodal displacements, the rotations, x and y.

The triangle has its vertices related to global coordinates at xi, yi, i = 1, 2 and 3 (only the
global directions are important, independently where the coordinates center is considered
for the elements). The triangle parameters contained in (4) are defined as

ak = −xij/L2ij bk = 3xijyij/4L2ij ck = (x2ij − 2y2ij)/4L2ij

dk = −yij/L2ij ek = (y2ij − 2x2ij)/4L2ij

with xij = xi − xj yij = yi − yj L2ij = x2ij + y2ij

and with k = 4, 5, 6 for the triangle sides 23, 31, 12, respectively.

The derivation of the 9 pairs of shape functions of (4) is given in appendix 1. Figure a1 of this
appendix gives also an explicit representation of the DKT triangle.

The differential equation of the deflection surface of a plate submitted to a transverse load
is given in function of the moments (Timoshenko and Woinowsky-Krieger, Theory of Plates
and Shells, article 21) by

d2Mx/dx2 + d2My/dy2 + 2d2Mxy/dxdy + f = 0

where f is the transverse load, and where the moments for a composite material without
coupling stiffness are expressed by {M} = [Db}.{K}:

{M} [Db] {K}

Mx D11 D12 D16 Kx
My = D12 D22 D26 Ky
Mxy D16 D26 D66 Kxy

with the vector of curvatures being here given by

5
 dx/dx

{K} = dy/dy

dx/dy + dy/dx

In order to derive the element matrix equation, the governing equation is recast in a
weighted-integral form:

∫.v.(d M /dx
2
x
2
+ d2My/dy2+ 2d2Mxy/dxdy + f).dA = 0

After trading twice the differentiation to the weight function, the left-hand side of the weak
form, which is used to establish the bending stiffness matrix, can be written as

∫
− d2v/dx2.Mx.dA − ∫d v/dy .M .dA
2 2
y ∫
− 2 d2v/dxdy.Mxy.dA

For the sake of brevity, we omit the rigidities D16 and D26, and so express the left-hand side
of the weak form by

∫ ∫ ∫
−D11 d2v/dx2.dx/dx.dA − D12 d2v/dx2.dy/dy.dA − D12 d2v/dy2.dx/dx.dA

−D22 ∫d v/dy .d /dy.dA − 2D ∫d v/dxdy.(d /dy + d /dx).dA

2 2
y 66
2
x y

In order to construct the 9 element equations that establish KeDKT, we substitute the
derivatives of the weight functions with the 9 pairs of shape functions defined in (4). These
pairs of shape functions imply subjacent, eventually immaterial (*), shape functions that
approximate the transverse displacement, w, and that would replace the weight function, v,
so that, as x = −dw/dx and y = −dw/dy, we make

(dv/dx)i = −(Hx)i (dv/dy)i = −(Hy)i for i = 1 to 9

Additionally, we substitute x and y with their approximations given in (3).

∫ ∫
(KeDKT)ij = D11 (dHx/dx)i.(dHx/dx)jdA + D12 (dHx/dx)i.(dHy/dy)j.dA +

∫ ∫
D12 (dHy/dy)i.(dHx/dx)j.dA + D22 (dHy/dy)i.(dHy/dy)jdA +

D66 ∫(dH /dy + dH /dx) .(dH /dy + dH /dx) dA

x y i x y j

is thus, omitting the rigidities D16 and D26, the expression to compute the entries of KeDKT in
cartesian coordinates.

(*) These shape functions are thus ignored in the DKT theory. In order to investigate the
consequences of this disregard, we will presume in appendix 3, “The obscure side of the DKT
element”, that they exist, and formulate the complete weak form, considering both sides of
it.

The strain-displacement matrix in cartesian coordinates being given by

6
 3x9
dHx/dx
[B(x,y)] = dHy/dy (5.a)
dHx/dy + dHy/dx

the KeDKT matrix can be written compactly in cartesian coordinates as

∫
KeDKT = [B(x, y)]T[Db][B(x,y)]dxdy (5.b)

However the shape functions have been expressed in area coordinates, and the DKT
element stiffness matrix in area coordinates would have the form (*)

∫ ∫
KeDKT = 2A 0 to 1 0 to 1 - [B( )]T[Db][B( )]dd (5.c)

where [B( )] is the strain-displacement matrix in the area coordinates system.

(*) We refer the reader to "The Finite Element Method, its Basis & Fundamentals" of
Zienkiewicz, Taylor & Zhu, 5.11.

Batoz et al. (1980) derives directly the KeDKT bending stiffness matrix from the expression of
the bending strain energy (appendix 2).

In order to establish matrix [B( )], we express the vector of curvatures, {K}, taking into
account the relations of (3)

(3 x 9) (9)
dHxT/dx {ue}

{K} ≈ dHyT/dy

dHxT/dy + dHyT/dx

It can be demonstrated that the derivatives of the H( ) with respect to the cartesian
coordinates can be expressed with respect to the area coordinates by the relation

dH( )/dx dH( )/d

= [J( )]
−1

dH( )/dy dH( )/d

with
y3 − y1 y1 − y2
[J( )] = −1
1/|J|
x3 − x1 x2 − x1

and [J] is the Jacobian matrix, whose determinant, |J|, is here equal to 2A.

Thus dH( )/dx = 1/2A.[y31.dH( )/d + y12.dH( )/d

7
 dH( )/dy = 1/2A.[−x31.dH( )/d − x12.dH( )/d

The vector of curvatures can be so expressed in function of the area coordinates by

{K} = [B( )].{ue} (6)

where
(3 x 9)
y31.dHx( )/d + y12.dHx( )/d

[B( )] = 1/2A −x31.dHy( )/d − x12.dHy( )/d

−x31.dHx( )/d − x12.dHx( )/d + y31.dHy( )/d + y12.dHy( )/d

If we derive with respect to the area coordinates ( ) the 9 pairs of functions contained in
each vector, {Hx} and {Hy}, to compute [B( )], the DKT stiffness matrix can be established
readily developing (5.c). The result of these derivations is given in the next page.

Because the integrals involve only quadratic terms, and when the thickness and the material
properties are constant over the element, the exact integration of KeDKT is obtained using
only three numerical integration points located at the mid-nodes. We refer the reader to
"The Finite Element Method, its Basis & Fundamentals" of Zienkiewicz, Taylor & Zhu, table
5.3:

  weights
1/2 1/2 1/3
1/2 0 1/3
0 1/2 1/3

In fact, as the triangle in area coordinates has an area of 1/2, and not 1, which is the sum of
the weights, the results of the weighted sum shall be here divided by 2.

8
 P6(1 − 2) + (P5 − P6)
q6(1 − 2) − (q5 + q6)
−4 + 6( + ) + r6(1 −2) − (r5 + r6)
−P6(1 − 2) + (P4 + P6)
dHx/d = q6(1 − 2) − (q6 − q4)
−2 + 6 + r6(1 −2) + (r4 − r6)
−(P5 + P4)
(q4 − q5)
−(r5 − r4)

−P5(1 − 2) − (P6 − P5)

q5(1 − 2) − (q5 + q6)
−4 + 6( + ) + r5(1 −2) − (r5 + r6)
(P4 + P5)
dHx/d = (q4 − q6)
−(r6 − r4)
P5(1 − 2) − (P4 + P5)
q5(1 − 2) + (q4 − q5)
−2 + 6 + r5(1 −2) + (r4 − r5)
(7)
t6(1 − 2) + (t5 − t6)
1 + r6(1 −2) − (r5 + r6)
−q6(1 − 2) + (q5 + q6)
−t6(1 − 2) + (t4 + t6)
dHy/d = −1 + r6(1 − 2) + (r4 − r6)
−q6(1 − 2) − (q4 − q6)
−(t4 + t5)
(r4 − r5)
−(q4 − q5)

−t5(1 − 2) − (t6 − t5)

1 + r5(1 −2) − (r5 + r6)
−q5(1 − 2) + (q5 + q6)
(t4 + t6)
dHy/d = (r4 − r6)
−(q4 − q6)
t5(1 − 2) − (t4 + t5)
−1 + r5(1 − 2) + (r4 − r5)
−q5(1 − 2) − (q4 − q5)

where Pk = −6xij/L2ij = 6ak tk = −6yij/L2ij = 6dk qk = 3xijyij/L2ij = 4bk rk = 3y2ij/L2ij

with k = 4, 5, 6 for the triangle sides 23, 31, 12, respectively.

9
The Load vector
When a uniform pressure load, q0, is applied on the element, the three corner nodes share
the load equally, so that the load vector would be given by

{fe}T = q0.A/3.[1 0 0 1 0 0 1 0 0] (8)

Post-processing
The three moments, Mx, My, and Mxy, collected in vector {M}, are calculated directly from
the relation

{M( )} = [Db].{K} = [Db].[B( )].{ue} (9)

As mentioned before, node 1 is at ( ) equals to (0, 0), node 2 at (1, 0) and node 3 at (0, 1).
Other points inside the element, or along its sides, can be determined by the relations

 = 1/2A[ y31.(x − x1) − x31.(y − y1)]

 = 1/2A[y12.(x − x1) − x12.(y − y1)]]

with 2A = x31.y12 − x12.y31

An explicit expression of the DKT stiffness matrix (without numerical

integration)
In his paper "An explicit formulation for an efficient triangular plate-bending element", Batoz
(1982) presents a new expression of the stiffness matrix of the DKT element that does not
require numerical integration. Although the author claims a significant reduction of algebraic
operations in the evaluation of the stiffness matrices and bending moments, which is true,
one specific local coordinates system shall be considered for each element, and additional
transformation operations from local to global coordinates, and vice-versa, are required. A
net advantage of this procedure over the previous is therefore questionable for plates. In
the case of shells, when any way coordinates conversions have to be done, the
computational economy of this explicit expression of KeDKT may however be justified. Our
own assessment is that both procedures are indeed very attractive for finite element
programs implementation on desktop and micro computers, however, for composite
materials, the previous procedure would be in most cases the more effective.

Following Batoz, we decompose the [B] matrix into the product of two matrices [L ] and []
of order 3 X 9 and 9 x 9, respectively, where [] is independent of  and :

[B( )] = 1/2A[L].[] (10)

with

1−−   0 0 0 0 0 0
[L] = 0 0 0 1−−   0 0 0
0 0 0 0 0 0 1−−  

The 9 x 9 matrix [] considers local coordinates with the element x-axis aligned with the
triangle side defined from node 1 to node 2, as shown in figure 1. Matrix [], which results
from the decomposition of matrix [B], is given below.

10
This decomposition of matrix [B] follows K. T. Joseph and K. S. Rao, "A fast algorithm for
triangular plate bending element", Znt. J. num. Meth. Engng, 14, 1100-1104 (1979).

3 (x 3 , y 3 )

(x 5 , y 5 )
(x 4 , y 4 )

1
(x 1 , y 1 ) = (0,0)

(x6, y 6 )
2 (x 2 , y 2 )
= (x 2 , 0)

Figure 1: Element's local coordinates

y3P6 0 −4y3 −y3P6

−y3P6 0 2y3 y3P6
y3P5 −y3q5 y3(2 − r5) y3P4
−x2t5 x23 + x2r5 −x2q5 0
[] = 0 x23 0 x2t4
x23t5 x23(1 − r5) x23q5 −x3t4
−x3P6 − x2P5 x2q5 + y3 −4x23 + x2r5 x3P6
−x23P6 y3 2x23 x23P6 + x2P4
x23P5 + y3t5 −x23q5 + (2 − r5)x23 −x3P4 + y3t4
(1 − r5)y3 + y3q5
Columns 5 to 9 follow next

0 −2y3 0 0 0
0 4y3 0 0 0
−y3(P4 +
y3q4 y3(r4 − 2) P5) y3(q4 − q5) y3(r4 − r5)
x3 0 x2t5 x2(r5 − 1) −x2q5
x3 + x2r4 −x2q4 −x2t4 x2(r4 − 1) −x2q4
−x23r5 −x3r4 −
x3(1 − r4) x3q4 −x23t5 + x3t4 x2 x3q4 + x23q5
−y3 2x3 x2P5 x2q5 (r5 − 2)x2
−y3 + x2q4 −4x3 + x2r4 −x2P4 x2q4 (r4 − 2)x2
−x23P5 + −x23q5 − −x23r5 − x3r4 +
(r4 − 1)y3 − (2 − r4)x3 − x3P4 − (t4 + x3q4 + (r4 − 4x2 + (q5 −
x3q4 y3q4 t5)y3 r5)y3 q4)y3

11
All coefficients in [] are the same as those given in (4) and (7), and have been defined
before.

Introducing equation (10) into the expression (5) for KeDKT, we obtain its explicit form

KeDKT = 1/2A[]T[DL].[] (11)

with ∫ ∫
[DL] = 1/2A 0 to 1 0 to 1 - [L]T[Db][L]dd (12)

and, where [Db] is the bending matrix in the local coordinates of the element. For composite
materials, matrix [Db] would have thus to be computed for each element with the material x-
axis aligned with the side 1-2 of the element; this requirement may obviously be fastidious
for most composite shells to be analyzed.

Due to the configuration of the [Db] and [L] matrices, matrix [DL] (9 x 9) can be computed for
orthotropic or isotropic materials without integration as

D11[R] D12[R]
[DL] = 1/2A D12[R] D22[R]
D66[R]

where we recall that, for isotropic materials,

D11 = D22 = Eh3/[12(1 − v2)] D12 = vD11 D66 = D11(1 − v)/2

Matrix [R] is given by

2 1 1
[R] = 1 2 1
1 1 2

The coordinates conversion required to build the global matrix equation follows the usual
transformation relation

[KeG] = [T]T[Ke][T]

where [KeG] and [Ke] are the stiffness matrices in global and local coordinates, respectively,
and

[T01]
[T] = [T01]
[T01]

In the case of plates, matrix [T01] is given by

1 0 0
[T01] = 0 Cos Sin
0 −Sin Cos

12
with  being counted counterclockwise from the global to the local x-axes.

In turn, the element nodal displacements in the local coordinates are obtained from the
global displacements by the relation

{ue} = [T]{ueG}

The vector {M}, which contains the three moments, Mx, My, and Mxy, is then obtained for an
element making

{M} = [Db][B( )]{ue} = 1/2A[Db][L][]{ue} (13)

Vector []{ue} is constant over the element, and matrix [Db][L] depends on  and , but is
particularly simple due to the leanness of both {Db] and [L]. Thus, the computation of {M} at
any points ( ) of an element is straightforward. The bending moments are not continuous
across contiguous elements.

13
B. Coupling in composite plates with the DKT bending element

Introduction
In the case of composites exhibiting a significative coupling stiffness, the in-plane forces and
bending moments shall be expressed in terms of the matrix equation

Nx A11 A12 A16 B11 B12 B16 x

Ny A12 A22 A26 B12 B22 B26 y
Nxy = A16 A26 A66 B16 B26 B66 xy
(B.1)
Mx B11 B12 B16 D11 D12 D16 x
My B12 B22 B26 D12 D22 D26 y
Mxy B16 B26 B66 D16 D26 D66 xy

It results that the differential equations, which express the equilibriums of the forces and
moments on an infinitesimal area of the plate, would require the consideration of the
coupling stiffnesses Bij when the corresponding weak forms are formulated. This is obvious
as the strains and curvatures in equation (B.1) above are expressed in terms of their
respective displacements in the plane elasticity and classical plate theories. When the
composite exhibits a relevant coupling stiffness, the merged element matrix equation will
thus take the form:

Bending
Shape
functions
Plane elasticity differential
equations & Weight [Kep]s [Kec]s {ueS} {FeS}
functions =
Bending differential
equations & Weight [Kec]b [Kep]b {ueB} {FeB}
functions
Plane Elast.
(B.2)
Shape funct.

where [Kep]s is the plane elasticity stiffness matrix and [Kep]b is the bending stiffness matrix,
here [KeDKT].

The Coupling Stiffness Matrix [Kec]s

Two coupling stiffness matrices, [Kec]b and [Kec]s, with entries expressed in terms of the
coupling stiffnesses Bij, will so appear in the element matrix equation. In order to construct a
coherent system of equations, these two matrices shall be derived, on one hand, with the
weight functions used in the plane elasticity theory to establish [Kec]s, or used in the bending
theory to establish [Kec]b. On the other hand, the shape functions shall be the same as those
used to express the displacements. These requirements lead to a hybrid system of weight
and shape functions as highlighted with colors in matrix equation (B.2). It can be shown that
[Kec]s comes out to be the transpose of [Kec]b, so that only one of these two matrices requires
to be developed. We will here develop the plane elasticity coupling stiffness matrix [Kec]s
consistent with the DKT bending element and the plane elasticity linear 6-dof triangular
element.

14
The governing equations of a plate underdoing small deformations in the (x, y) plane are
easily derived as

dNx/dx + dNxy/dy = 0

dNxy/dx + dNy/dy = 0

Suitable weak forms to deduce the plane elasticity coupling stiffness matrix, [Kec]s, are then
developed as follows.

First differential equation:

∫ v.dN /dx.dA
el x + ∫ v.dN
el xy/dy.dA =0

In order to trade the differentiation with the weight function, we use the two following
identities.

d/dx(v.Nx) = v.dNx/dx + dv/dx.Nx and d/dy(v.Nxy) = v.dNxy/dy + dv/dy.Nxy

Then, we write the weak form for the first differential equation as

∫ dv/dx.N .dA
el x + ∫ dv/dy.N
el xy.dA = ∫ v.N .dy + ∫ v.N
el x el xy.dx (B.3a)

Second differential equation:

Proceeding identically for the second differential equation we get

∫ dv/dx.N
el xy.dA + ∫ dv/dy.N .dA
el y = ∫ v.N
el xy.dy + ∫ v.N .dx
el y (B.3b)

We introduce now in the left-hand side integrals of (B.3a) and (B.3b) the coupling terms for
Nx, Ny and Nxy implied in (B.1), and obtain at once the following terms needed to build the
plane elasticity coupling stiffness matrices, [Kec]s.

Terms from the first differential equation

∫
B11 eldv/dx.dx/dx.dA

∫
B12 eldv/dx.dy/dy.dA

∫
B66 eldv/dy.(dx/dy + dy/dx).dA
(B.4a)
∫
B26 eldv/dy.dy/dy.dA

∫
B16 eldv/dy.dx/dx.dA

∫
B16 eldv/dx.(dx/dy + dy/dx).dA

15
 Terms from the 2nd differential equation
∫
B12 eldv/dy.dx/dx.dA

∫
B22 eldv/dy.dy/dy.dA

∫
B66 eldv/dx.(dx/dy + dy/dx).dA
(B.4b)
∫
B16 eldv/dx.dx/dx.dA

∫
B26 eldv/dx.dy/dy.dA

∫
B26 eldv/dy.(dx/dy + dy/dx).dA

Shape and weight functions

The weight functions and their numbers shall be obviously the same that are used for the
plane elasticity linear element, while the displacements shall be approximated using the
same shape functions we have been using for the DKT bending element. Consequently, in
order to establish the plane elasticity coupling stiffness matrix for the DKT 5-dof per node
plate element, we shall use Lagrange linear functions for weight functions and the shape
functions Hx and Hy to approximate x and y, respectively.

These integrals in (B.4a) and (B.4b) will form the plane elasticity coupling stiffness matrix
[Kec]s. The subscripts c and s are here used to refer to the coupling terms in the plane
elasticity (extensional) differential equations.

For the linear plane elasticity element, the element matrix equation considers the three
Lagrange shape functions, L1, L2 and L3:

u = L1.u1 + L2.u2 + L3.u3

(B.5a)
v = L1.v1 + L2.v2 + L3.v3

where the linear Lagrange shape functions are given by, {L} = [L1 L2 L3]T, :

x2y3 − x3y2 −(y3 − y2) x3 − x2 1

{L} = 1/2A −(x1y3 − x3y1) y3 − y1 −(x3 − x1) x (B.5b)
x1y2 − x2y1 −(y2 − y1) x2 − x1 y

Using the three Lagrange linear functions as weight functions and substituting x and y with
their expression given in (3) in each of the two sets of weak terms (B.4a) and (B.4b), we get
the general terms for the submatrices [Kec(1)]s and [Kec(2)]s, respectively, which compose the
extensional coupling stiffness matrix [Kec]s:

∫∫
[Kec(1)]s = (B11.dLi/dx.(dHx/dx)j + B12.dLi/dx.(dHy/dy)j + B66.dLi/dy.(dHx/dy + dHy/dx)j +
B26.dLi/dy.(dHy/dy)j + B16.dLi/dy.(dHxj/dx)j + B16.dLi/dx.(dHx/dy + dHy/dx)j).dxdy (B.6a)

16
 ∫∫
[Kec(2)]s = (B12.dLi/dy.(dHx/dx)j + B22.dLi/dy.(dHy/dy)j + B66.dLi/dx.(dHx/dy + dHy/dx)j +
B16.dLi/dx.(dHx/dx)j + B26.dLi/dx.(dHy/dy)j + B26.dLi/dy.(dHx/dy + dHy/dx)j).dxdy (B.6b)

These two matrices have each 3 rows and 9 columns (i = 1 to 3; j = 1 to 9) and add to each
other vertically to form the extensional coupling matrix [Kec]s:

[Kec(1)]s
[Kec]s = (3 x 9)
(B.6c)
6 x 9) [Kec(2)]s
(3 x 9)

The plane elasticity coupling stiffness matrix attaches to the extensional stiffness matrix
[Kep]s, to form an underdetermined system of 6 equations with 15 unknowns:

[Ke]s{ue} = {fe}s + {Qe}s (B.7a)

where

[Ke]s = [Kep]s [Kec]s (B.7b)

(6x15) (6x6) (6x9)

and {ue} is the vector of 15 entries containing all nodal displacements. {ue} is written
explicitly below.

{ues}
{ue} = (B.7c)
{ueb}

{ues} = [u1, u2, u3, v1, v2, v3]T

{ueb} = [w1, x1, y1, w2, x2, y2, w3, x3, y3]T

In terms of cartesian coordinates, the two submatrices [Ke(c1)]s and [Ke(c2)]s can be written in
matrical notations as

∫
[Ke(c1)]s = [dL(x, y)][B(c1)][B(x,y)]dxdy (B.8a)

∫
[Ke(c2)]s = [dL(x, y)][B(c2)][B(x,y)]dxdy (B.8b)

where [B(x, y)] is the strain-displacement matrix defined before in (5.a), and

[B(c1)] (2 x 3) [B(c2)] (2 x 3)
B11 B12 B16 B16 B26 B66
B16 B26 B66 B12 B22 B26

[dL(x,y)] (3 x 2)
dL1/dx dL1/dy
dL2/dx dL2/dy
dL3/dx dL3/dy

17
In terms of area coordinates, we would have

∫ ∫
[Ke(c1)]s = 2A 0 to 1 0 to 1 - [dL( )][B(c1)][B( )]dd (B.9a)

∫ ∫
[Ke(c2)]s = 2A 0 to 1 0 to 1 - [dL( )][B(c2)][B( )]dd (B.9b)

where [B( )] has been defined before in (6) and [dL( )] is given below in (B.10).

The linear Lagrange shape functions in area coordinates are given by

L1 = 1 −  −  L2 =  L3 = 

and, as before, we have

dLi( )/dx = 1/2A.[y31.dLi( )/d + y12.dLi( )/d

dLi( )/dy = 1/2A.[−x31.dLi( )/d − x12.dLi( )/d

so that

−(y31 + y12) x31 + x12

[dL( )] = 1/2A y31 −x31 (B.10)
y12 −x12

Using expressions (B.9a) and (B.9b), the extensional coupling matrix [Ke(c)]s can be easily
implemented on a mini- or micro-computer with, again, three numerical integration points
located at the mid-nodes. The bending coupling matrix [Ke(c)]b can be then established as the
transpose of [Ke(c)]s.

The Extensional Stiffness Matrix [Kep]s

With the element coordinates center at the centroid of the triangle, the element linear
generalized dof extensional matrix takes the simple expression:

A11 A16 A16 A12

A16 A66 A66 A26
[KeDp]s = A (B.11a)

A16 A66 A66 A26

A12 A26 A26 A22

The extensional stiffness matrix is then established making

[Kep]s = [hL2−T][KeDp]s[hL2−1] (B.11b)

[hL2]
1 x1 y1 [hL]
[hL] = 1 x2 y2 [hL]
1 x3 y3

18
The Element Coupled Matrix
The element coupled matrix equation with 15 degrees of freedom can be now assembled as
shown below.

where {ues} and (ueb} have been defined before in (B.7c). Using the subscript cc for "coupled
& consolidated", we write equation (B.12a) succinctly as

[Ke]cc{ue} = {Fe}cc (B.12b)

where {Fe}cc = {fe}cc + {Qe}cc

Global matrix equation assembly

In order to assemble the finite element global equation, it is required to perform first a trade
between rows and between columns in [Ke]cc, and between rows in {ue} and {fe}cc. The
purpose is to have the 5 edofs, ui, vi, wi, (x)i and (y)i together at each node. This is however
a straightforward and simple operation, which is shown in the next page for vector {ue}.

The same rows trade should be performed in [Ke]cc bringing row 4 to the position of row 2,
row 7 to the position of row 3, and so on. In addition, it is required to swap afterwards the
columns of [Ke]cc in the same fashion it was done for the rows, bringing column 4 to the
position of column 2, column 7 to the position of column 3, and so on.

19
 From row to row
1 u1 u1 1 1
2 u2 v1 4 2
3 u3 w1 Node 1 7 3
4 v1 (x)1 8 4
5 v2 (y)1 9 5
6 v3 u2 2 6
7 w1 v2 5 7
8 (x)1 w2 Node 2 10 8
==> ==>
9 (y)1 (x)2 11 9
10 w2 (y)2 12 10
11 (x)2 u3 3 11
12 (y)2 v3 6 12
13 w3 w3 Node 3 13 13
14 (x)3 (x)3 14 14
15 (y)3 (y)3 15 15
(B.13)

Numerical results of angle-ply laminates

Expressive coupling stiffnesses, B16 and B26, surge in antisymmetric angle-ply laminates with
few layers. We have so analyzed (−45/45) two plies laminates, which are thus able to reveal
the importance of the coupling effect in composites, and to evaluate the DKT 15-dof
composite plate element here presented. Same exercise has been performed for (−45/45)4
laminates alternating −45° and +45° on the eight layers so that much lower coupling stiffness
is produced. The coupling effect so appears when we compare the results of the two
laminates, with two and eight plies, respectively.

J. N. Reddy in its “Mechanics of Laminated Composite Plates and Shells” performed originally
the same exercise. Reddy's table 9.2.3 compares results obtained by FEA based on the
classical plate theory, and rectangular 20-dof elements, with the analytical results obtained
with the Navier solution. The exercise considers the following laminae characteristics

E1 = 25E2 G12 = 0,5E2 v12 = 0,25

He supplies results for the following nondimensionalized plate depression and stresses at
specified locations:

w = w0(0,0).E2h3/(B4.q0) x 100
x = x(0, 0, h/2). (h/A)2(1/q0)
y = y(0, 0, h/2). (h/A)2(1/q0)
xy = xy(A/2, B/2, -h/2). (h/A)2(1/q0)

where (0, 0) is the laminate center, (A/2, B/2) its corner, h its thickness, and q0 is the
uniformly distributed load. Square laminates (A = B) are here considered.

The exercise gives thus a comparison of finite element solutions with the analytical solutions
of simply supported angle-ply square laminates under a uniformly distributed transverse

20
load. In all finite element analyses, a quadrant of the plate was used. Reddy explains that
antisymmetric angle-ply laminates admit Navier solutions only for simply supported SS-2
boundary conditions. These conditions are therefore considered for all cases here reported.
The symmetry conditions can be identified from the Navier solution, and are given below
along with the SS-2 boundary conditions.

SS-2 at x = ± A/2 at y = ± B/2

boundary u0 = w0 = x = 0 v0 = w0 = y = 0

SS-2 at x = 0 at y = 0
symmetry v0 = y = 0 u0 = x = 0

The calculation of the stresses requires first to determine the mid-plane strains from the
element nodal displacements, ui and vi. The mid-plane displacements are here approximated
by linear functions of coordinates, x and y, related to reference axes situated at the centroid
of the triangle:

u = D1 + D2x + D3y
v = D4 + D5x + D6y

The generalized dof coefficients, D1 to D6, collected in vector {De}, are obtained directly from
the element nodal displacements making

{De} = [hL2−1].{ues}

where (ues} has been given in (5.7d), and {De} = [D1, D2, D3, D4, D5, D6]T

We have so directly the relations that give the mid-plane strains, assumed here constant
over the element (so that this element is known as the Constant Strain Triangle – CST):

0x = du/dx = D2
0y = dv/dy = D6
0xy = du/dy + dv/dx = D3 + D5

The total strains, considering altogether bending and extensional deformations, are then
given by

x = 0x + z.dx/dx
y = 0y + z.dy/dy
xy = 0xy + z.(dx/dy + dy/dx)

where z is the distance of the considered lamina to the mid-plane of the plate, and where
the curvatures are obtained, as before for the DKT bending element, by equation (6):

{K} = [B( )].{ueb}

Finally, we determine the stresses in the laminate coordinates applying to the strains the
stiffness matrix [Q]xy pertaining to any selected lamina. We refer the reader, who is not
acquainted yet with the analysis of composite laminates, to Contrapunctus VIII, or to Daniel
& Ishai "Engineering Mechanics of Composite Materials", or to the already mentioned
Reddy's book.

21
 x = Qxx.x + Qxy.y + Qsx.xy
y = Qyx.x + Qyy.y + Qys.xy
xy = Qsx.x + Qsyy.y + Qss.xy

The nondimensionalized stresses, at the locations given before, were so calculated for the
DKT 15-dof composite plate element and compared to

(1) The Navier analytical solutions.

(2) The FEA solutions obtained by Reddy with the CPT 4-node, 5-dof per node, rectangular,
nonconforming, element usually denoted as CPT(N). In this case, Reddy evaluated the
stresses at the Gauss point nearest to the specified location. For this nonconforming
rectangular plate element, the strains and stresses were evaluated using the one-point
Gauss rule, i.e., the center of the rectangular element.
(3) The FEA solutions obtained with the triangular element based on the already mentioned
Kirchhoff triangular bending element, coupled with the linear plane elasticity element,
and referred here also as CPT(N). In this case, the stresses require to be evaluated at the
element centroid.

Results are given in the two following tables. For all finite element analyses, the quarter
plate was divided according to the information given in the tables, and mesh B (see below in
figure 6) was used for all triangular elements.

Table B.1: Angle-Ply (−45°/45°) - 2 plies - Plate simply supported and under a uniformly distributed load
Source Procedure Mesh w xx yy xy
Contrapunctus XVI CPT[N] 3-node 4x4 1,0759 0,3022 0,3022 0,4222
Reddy table 9.2.3 CPT[N] 4-node 4x4 1,0341 0,3279 0,3279 0,4264
Present paper DKT 3-node 4x4 1,0090 0,3690 0,3664 0,4116
Present paper DKT 3-node 8x8 1,0236 0,3555 0,3549 0,4342
Present paper DKT 3-node 12x12 1,0261 0,3529 0,3526 0,4389
Present paper DKT 3-node 16x16 1,0270 0,3519 0,3517 0,4407
Present paper DKT 3-node 20x20 1,0274 0,3514 0,3513 0,4415
Reddy table 9.2.3 Analytical − 1,0280 0,3504 0,3504 0,4421

Table B.2: Angle-Ply (−45°/45°)4 - 8 plies - Plate simply supported and under a uniformly distributed load
Source Procedure Mesh w xx yy xy
Contrapunctus XVI CPT[N] 3-node 4x4 0,4541 0,1741 0,1741 0,1785
Reddy table 9.2.3 CPT[N] 4-node 4x4 0,3905 0,1880 0,1880 0,2612
Present paper DKT 3-node 4x4 0,3868 0,2078 0,2076 0,2460
Present paper DKT 3-node 8x8 0,3861 0,2023 0,2022 0,2586
Present paper DKT 3-node 12x12 0,3859 0,2013 0,2013 0,2615
Present paper DKT 3-node 16x16 0,3859 0,2010 0,2010 0,2626
Present paper DKT 3-node 20x20 0,3858 0,2009 0,2009 0,2632
Reddy table 9.2.3 Analytical − 0,3858 0,2006 0,2006 0,2637

The convergence of the DKT 15-dof element is shown in the two following figures. The
results are already within ±2% of target with an 8x8 mesh, and at less than ±1% with a 12x12
mesh. In all cases, the DKT 15-dof element converges very accurately towards the analytical
result.

22
The results, obtained with different elements and a coarse mesh of 4x4, are not quite
comparable because the stresses were computed at different locations. Even so, we may
conclude that the DKT 15-dof element performs at least comparably with the CPT(N)
rectangular element reported by Reddy. The DKT element allows evaluating the strains and
stresses at any specified point within the element, and so its results could be compared
directly with the analytical results to assemble the convergence charts. The CPT(N) 4-node
element is a robust and well-accepted element, it produces here good results considering
the coarse mesh and the location considered for the stresses. The CPT(N) 3-node element, in
congruence with the several restrictions commented and documented in this paper,
performs marginally in this exercise.

Figure 2

Figure 3

23
C. Buckling of composite plates with a consistent geometrical stiffness
matrix

Introduction
“As far as the geometric stiffness matrix is concerned, inconsistent forms have been often
adopted by assuming ad hoc shape functions for w within the element (Bathe and Ho, 1981;
Mateus et al., 1997; Doyle, 2001; Khosravi et al., 2007). If the modified approximations
obtained for x and y after enforcing the Kirchhoff constraint at selected points are used to
generate the geometric stiffness matrix, such a matrix is said to be consistent because their
shape functions are the same used in establishing the bending stiffness matrix.”
"An Explicit Consistent Geometric Stiffness Matrix for the DKT Element", Lucena Neto et al., Instituto Tecnológico
de Aeronáutica, São José dos Campos, Brazil (2017), www.lajss.or

The Geometrical Stiffness Matrix

If in addition to transverse loads, there are forces acting in the middle plane of the plate,
these forces may have a considerable effect on the bending of the plate and must be
considered in the differential equation of the deflected surface. When therefore we consider
the contribution of in-plane forces, Nx, Ny and Nxy, to the transverse deflection of the plate,
w, the differential equation of the deflection surface of a plate submitted to a transverse
load, f, becomes (Timoshenko and Woinowsky-Krieger, Theory of Plates and Shells, article
90)

d2Mx/dx2 + d2My/dy2 + 2d2Mxy/dxdy

− Nx.d2w/dx2 − Ny.d2w/dy2 − 2Nxy.d2w/dxdy + f = 0 (C.1)

The consideration of the in-plane forces in equation (6.1) gives rise to the geometrical
stiffness, or stability matrix, [GeDKT] , which shall be added to the stiffness matrix [KeDKT]
established before. Depending on having in-plane forces in tension or compression, the
element will become stiffer or more flexible, respectively, so that the deflections will
become correspondingly smaller or larger than those of the plate bent by transverse load
only. Such geometrical stiffness matrix is established recasting the in-plane forces terms of
(C.1) in a weighted-integral form:

− ∫el.v.(N .d w/dx
x
2 2
+ 2Nxy.d2w/dxdy + Ny.d2wdy2).dA (C.2)

After trading the differentiation to the weight function, v, one obtains the weak form that
originates the geometrical stiffness matrix:

∫ N .dv/dx.dw/dx.dA + ∫ N .dv/dy.dw/dy.dA
el x el y

+ ∫ N .(dv/dx.dw/dy + dv/dy.dw/dx).dA
el xy (C.3)

Proceeding as before, we substitute the derivatives of the weight functions with the 9 pairs
of shape functions defined in (4), and we replace the derivatives of w with their
approximations given in (3):

(dv/dx)i = −(HTx)i (dv/dy)i = −(HTy)i for i = 1 to 9

(C.4)
dw/dx = −x = −HTx.{ue} dw/dy = −y = −HTy.{ue}

24
so that
∫ ∫ ∫
(GeDKT)ij = Nx {Hx}i.{Hx}jdA + Ny {Hy}i.(Hy)j.dA + Nxy [{Hx}i.{Hy}j + {Hy}i.{Hx}j}dA
(C.5)

is the expression to compute the entries of GeDKT in cartesian coordinates.

In area coordinates, the geometrical stiffness matrix is so expressed by

∫ ∫
GeDKT = 2A 0 to 1 0 to 1 - [H( )][N][H( )]Tdd (C.5)

where we define matrices [H] and [N] as

[H] (9 x 2) [N] (2 x 2)
{H}x {H}y Nx Nxy
Nxy Ny

As there is no numerical integration available for quartic orders, the exact integration of the
element geometric stiffness would require 7 integration points, which consider quintic
orders. However, it is observed that the 3 integration points, used before, maintains correct
rank and so, due to its greater economy, would be also suitable in practice. The 7 points
integration scheme, extracted from "The Finite Element Method, its Basis & Fundamentals"
of Zienkiewicz, Taylor & Zhu, table 5.3, is given below.

  weights
1/3 1/3 0,225 1 = 0,059715872
1 1 0,132394153 1 = 0,470142064
1 1 0,132394153 2 = 0,797426985
1 1 0,132394153 2 = 0,101286507
2 2 0,125939181
2 2 0,125939181
2 2 0,125939181

When the combined action of transverse and in-plane forces are considered, the element
bending matrix equation becomes

[[KeDKT] + [GeDKT]].{ue} = {fe} + {Qe} (C.7)

Here, {Qe} incorporates shear forces and moments concentrated at the nodes acting on the
sides of the elements; they arise from both transverse and in-plane loads. Upon global
matrix equation assembly, the vectors {Qe} make up {QG}, whose entries provide the reactive
shearing forces and bending moments on the boundary of the plate, and are equal to zero
elsewhere.

Numerical results with angle-ply laminates

We consider now rectangular plates submitted to a virtual state of loading where in-plane
forces by unit length, N*x, N*y and N*xy act uniformly over the plate. This virtual state of
loading allows analytical solutions for some types of boundary conditions and for some
types of laminates, and, therefore, is worthy to test the reliability of finite element models

25
designed for composites. In his "Mechanics of Laminated Composite Plates and Shells,
Chapter 6", Reddy offers comprehensive examples where he represents these forces as
uniformly distributed along the edges of the plates as shown in figure 4. Reddy treats of
uniaxial (N*x) and biaxial (N*x, N*y = kN*x) loadings.

Figure 4

Within the classical laminated plates theory, Navier solutions can be developed for the
buckling of rectangular composite laminates with two sets of simply supported boundary
conditions (Reddy, Mechanics of Laminated Composite Plates and Shells, 5.5.2: The Navier
Solution):

at x = 0 and x = A at y = 0 and y = B
v0 = w0 = dw0/dy = 0 u0 = w0 = dw0/dx = 0
SS-1
Nx = Mx =0 Ny = My =0

u0 = w0 = dw0/dy = 0 v0 = w0 = dw0/dx = 0
SS-2
Nxy = Mx =0 Nxy = My =0

In the Navier method, the displacements are expanded in a double trigonometric series. The
choice of the functions in the series is restricted to those that satisfy the boundary
conditions. The Navier solutions with SS-1 boundary conditions can be found only for
laminates whose stiffnesses A16, A26, B16, B26, D16 and D26 are zero. Furthermore, Navier
solutions with SS-2 boundary conditions can be found only for laminates whose stiffnesses
A16, A26, B11, B12, B22, B66, D16 and D26 are zero. Antisymmetric angle-ply laminates will
therefore satisfy the criteria enabling a Navier solution for SS-2 boundary conditions, while
antisymmetric cross-ply laminates will satisfy the criteria for SS-1 boundary conditions.
Laminates with a single orthotropic layer or symmetrically laminated plates with multiple
specially orthotropic (*) layers would satisfy the criteria for both boundary conditions.

(*): A specially orthotropic layer is a layer whose principal material coordinates coincide with
those of the plate.

We here limit ourselves to treat with DKT elements antisymmetric angle-ply laminates, and
compare our results with Navier solutions as informed in table 6.4.2 of Reddy's book. The
table contains nondimensionalized buckling loads for simply supported (SS-2) antisymmetric
angle-ply rectangular laminates under uniaxial and biaxial edge forces versus different plate
aspect ratio, A/B. The following properties of the laminae are considered.

E1/E2 (varied) G12 = 0,5E2 v12 = 0,25

26
Analytical nondimensionalized buckling loads found for (45/-45)k (k = 1 and 4) laminates are
informed in Reddy's table 6.4.2 and transcribed in tables C.1 and C.2 below.

(45/−45) Nondimensionalized Buckling loads = Ncr.B2/(E2h3)

E1/E2: 10 20 25 40
Uniaxial Compression (k = 0)
0,5 12,633 18,140 20,825 28,809
A/B 1 9,060 13,373 15,475 21,713
1,5 9,603 23,963 16,285 22,779
Biaxial Compression (k = 1)
0,5 11,893 14,518 16,660 23,045
A/B 1 4,530 6,692 7,738 10,856
1,5 3,129 9,021 5,270 7,353
Table C.1 (Analytical references)

(45/−45)4 Nondimensionalized Buckling loads = Ncr.B2/(E2h3)

E1/E2: 10 20 25 40
Uniaxial Compression (k = 0)
0,5 23,746 43,841 53,888 84,020
A/B 1 17,637 33,320 41,166 64,685
1,5 18,565 34,909 43,091 67,607
Biaxial Compression (k = 1)
0,5 18,999 35,076 43,110 67,222
A/B 1 8,813 16,660 20,578 32,343
1,5 6,001 11,251 13,877 21,743
Table C.2 (Analytical references)

Finite element analyses with DKT elements were performed for the four cases in red in the
tables, that is, for uniaxial compression (k = 0), plate aspect ratio of 1 (A = B), and laminates
with two and eight layers. Results are compared with the analytical results and those that
we obtained with 20-dof rectangular elements. In all finite element analyses, a quarter plate
was divided in 4 x 4 elements.

Nondimensionalized Buckling loads = Ncr.B2/(E2h3)

Uniaxial Compression (k = 0) E1/E2
Source Procedure 20 25
Reddy Analytical 13,373 15,475
(45/−45) This paper DKT 13,640 15,800
Contrapunctus XV CPT 4-node 13,302 15,396
Reddy Analytical 33,320 41,166
(45/−45)4 This paper DKT 33,246 41,068
Contrapunctus XV CPT 4-node 32,902 40,651
Table 6.3

The agreement among the results is very good. The bending-stretching coupling reduces
severely the buckling load for the two-layer plate. Reddy informs that the buckling load is
the maximum for  = 45°. The difference verified between the two numerical integrations,
with three or seven points, deserves to be illustrated: although the same critical buckling

27
loads were obtained in all cases, they are approached differently. This is shown in figure 5
where the central deflection of the plate was plotted against the nondimensional axial load,
Nx, for the two-layer laminate with a modulus ratio of E1/E2 = 20. The iterations were
realized with a step value of Nx ≈ 0,005. Also, a slight transverse load requires to be
applied on the plate in order to reveal clearly the buckling load in the finite element
calculations.

Figure 5

28
D. Evaluation of the DKT element in shell benchmarks

Introduction
The two papers, "A study of three-node triangular plate bending elements", Batoz et al.
(1980) and "An explicit formulation for an efficient triangular plate-bending element", Batoz
(1982) contains several numerical examples where the DKT element compares favorably
against other elements. Follows a list of the numerical examples presented in the two
papers.

Batoz et al (1980)
• Square plate under concentrated and uniform loads with clamped and simply
supported edges.
• Twisting of a square plate.
• Static analysis of a rhombic cantilevered plate subjected to a uniform load
• Free vibration analysis of a square plate with clamped and simply-supported edges
• Free vibration analysis of a triangular cantilever.

The authors conclude: "the discrete Kirchhoff theory (DKT) and hybrids stress model (HSM)
elements have been found to be the most effective 9-dof triangular elements available for
bending analysis of thin plates. Both elements can be formulated and implemented very
efficiently. Their effectiveness has been observed in a variety of static and dynamic analyses.
Based on theoretical and computational considerations and the numerical results, the DKT
element is found to be somewhat superior to the HSM element. Because of little storage
requirements, these two elements appear to be very promising for structural analysis on
mini- and micro-computers."

Batoz (1982)
• Patch test
• Behavior of a cantilever rectangular plate under twisting load
• Convergence analysis of clamped and simply-supported rectangular plates (with
aspect ratios of 1, 2, 3) subjected to a central concentrated load

The author concludes: "the results presented in this paper, together with the previous results
published in Reference 1 (Batoz et al. 1980), confirm the excellent behavior of the DKT
element compared to the other 9-dof plate-bending triangular elements."

We just verified that the evaluation of the DKT element on shells is out of the scope of the
two mentioned papers; also, It has been recognized that a small set of linear test cases
would be critical to evaluate shell elements (see, for example, the survey paper by
Belytschko, 1986). The set contains the three cases we have been using to test our shell
elements: the Scordelis-Lo roof, the pinched cylinder, and the hemispherical shell with point
loads. We will so hereafter compare the so built DKT shell element with other 18-dof
triangular flat shell elements on these three famous benchmarks, which we will not describe
again as they have been already described in Contrapunctus XXI. Furthermore, there are
plenty of references on these tests in the literature and on the Internet for whom is still not
familiar with them.

The Drilling Degrees of Freedom

An 18-dof DKT shell element can be built merging the herein 9 x 9 bending stiffness matrix
with the 6 x 6 plane elasticity matrix, as we did in the previous section. The so obtained

29
three translations (u, v, w) and two rotations (x, y), about the x- and y-axes, are then
completed with the "drilling" degree of freedom, which is the rotation about the z-axis
normal to the plane of the element. Considering the element matrix equation given in the
previous section after rows and columns swap for the global matrix equation assembly, the
drilling stiffnesses would be introduced adding sixth, twelfth and eighteenth rows and
columns in the plate stiffness matrix. The drilling stiffnesses will be so placed in the nine
cells, (6, 6), (6, 12), (6, 18), (12, 6), (12, 12), (12, 18), (18, 6), (18, 12), and (18, 18).
Zienkiewicz and Taylor (*) propose to add a unit stiffness to the diagonal of each drilling
degree of freedom and subtract 1/(n-1) (n being the number of nodes) of that stiffness from
each off-diagonal drilling degree of freedom in the same row and column. This retains
symmetry, equilibrium and freedom from forces under rigid body motion. A "basic" stiffness
taken equal to 4D66/√3 would be adequate in most applications for triangular elements. The
nine drilling stiffnesses would be so obtained for a triangular element from the drilling
stiffness matrix, [Kez], given below.

1 −1/2 −1/2
e
[K z] = kz −1/2 1 −1/2
−1/2 −1/2 1

with kz = 4D66/√3

(*) Finite Element Method: Volume 2, Fifth Edition by O. C. Zienkiewicz and R. L. Taylor

The transposition matrix, which is used to transpose the elements equation matrices to a
common global coordinates system, is composed of an elemental cosines matrix repeated
2n times, where n is the element number of nodes. For a 3-node triangular element, we
have thus the following transposition matrix:

T06
T01
T01
T01
T01
T01
T01

T01
Cos(x'x) Cos(x'y) Cos(x'z)
Cos(y'x) Cos(y'y) Cos(y'z)
Cos(z'x) Cos(z'y) Cos(z'z)

where (x'x) is the angle measured from the element x-axis to the global x-axis, (x'y) is the
angle measured from the element x-axis to the global y-axis, and so on. This transposition
matrix is called appropriately the direction cosines matrix as only the directions of the axes
are important, not mattering where, in the space, eventual centers of coordinates are
considered.

The inverse of the transposition matrix is equal to its transpose, [T06]-1 = [T06]T, and

[Ke]G = [T06]T[Ke]L[T06]

30
where [Ke]G and [Ke]L are the element stiffness matrices in the global and local coordinates,
respectively. [Ke]L is the element stiffness matrix where, for global matrix assembly, the rows
and columns swaps were effected in [Ke]cc in order to have the 3 translations and the 3
rotations together at the nodes in the sequence given in (5.13) adding (z) as a sixth nodal
displacement. For the load vector, we have

[fe]G = [T06]T[fe]L

The mesh configurations used in our analyses are illustrated in figure 6. We have so
reticulated the shell surface in quadrangles, then, different triangular mesh patterns,
meshes "A" and "B", have been generated slanting different ways one diagonal of these
quadrangles. An additional mesh pattern, which considers both diagonals, and that we have
called mesh "D", has also been used. Mesh "D" considers an additional node in the center of
the quadrangular cell, and, thus, requires more dofs in the global matrix equation.

Figure 6

Mesh "A" performs generally better than mesh "B", and is preferentially used. However,
mesh "B" is required, in most cases, due to singularity, for the classical 3-node Kirchhoff
triangular plate element mentioned before; and, we will refer to the shell element built from
it as the CPT element. Mesh "D", which comprises the two simple meshes, "A" and "B",
provides equilibrium to the cell, and, produces normally the best accuracy and the fastest
convergence. Besides the CPT element, the following two other 18-dof triangular flat shell
elements will be compared with the DKT flat shell element:

• The TRIC element of Teneketzis and Argyris described in their book "Finite Element
Analysis for Composite Structures", whose fundaments have been summarized in
Contrapunctus XI.
• The Utku FSDT element that we have developed in Contrapunctus XX based on an
argument of Prof. Senol Utku.

The TRIC and the Utku elements consider shear deformation, while the CPT and DKT
elements consider thin shells and are bound to the Kirchhoff constraints.

The Scordelis-Lo Roof

The displacement at the middle of the free edge of the roof is expected to be wref = −3.6288
in. = −0.09217 m; the convergences to this value with mesh "A", or "B" for the CPT element,
and dividing equally the hoop and the span of the roof, are shown in figure 7 for the four
elements. Due to symmetry only one quadrant of the shell requires to be discretized.

A comment is in order about the convergence target of the Scordelis-Lo roof. Obviously, the
transition for plates or shells from thick to thin is not clear-cut, so that exists, in
"moderately" thin plates or shells, a range of aspect ratios where will be observed a small
shear deformation added to bending. This is the case of the Scordelis-Lo roof, which has a
relative high aspect ratio of 200 when we consider the span of the roof, but with an aspect

31
ratio of only 100 related to its radius. From extensive simulations with different elements
and mesh designs, the added shear subsidence on the middle of the roof free edge would be
around two percent; this is what is observed in figure 7 when we compare the convergence
of the DKT with the TRIC element.

We observe also in figure 7 that no element has reached a complete convergence with the
finest mesh (30x30) that was tested. The rate of convergence can be, however, dramatically
improved reducing the cells skewness, for example, with a (2nx3n) reticulation, and
resorting to mesh D.

Figure 7

The Pinched Cylinder

The depression of the shell under the load point is expected to be wmax = −1.8541 x 10-5,
Flügge (1973). The convergence of the 4 elements toward this value is shown in figure 8. The
accurate determination of the depression under a load point is a serious problem due to the
produced sharp inflection in the shell surface, and mesh "D" has been so here preferentially
used, except for the CPT element, which requires mesh "B". Also, with a (3nx2n)
reticulation, the surface of the shell has been divided so that approximate square cells are
generated. Due to symmetry only one quadrant of the shell requires to be discretized.

32
 Figure 8

The Hemispherical Shell with Point Loads

In order to accommodate quadrilateral elements, the problem is often modeled with a hole
at the top of the hemisphere (MacNeal and Harder, 1985). We will here follow the original
work (Morley and Moris, 1978), without hole at the top. The problem only requires one-
quarter shell to be discretized.

The quarter shell has been reticulated (nxn) dividing it with equal numbers of parallels and
meridians. Mesh "A" has been used for the DKT, TRIC and Utku elements. The CPT element
will not perform adequately in this exercise due to the occurrence of singularities in some of
the cells. The deflection under the loads is expected to be equal to 0,0924; and figure 9
shows us the convergence of the three tested elements toward this value.

From the results of our evaluation on these three cases, we conclude that the TRIC and DKT
elements perform quite similarly in terms of accuracy and speed of convergence. As already
commented, the use of the CPT element is severely restricted to a few specific mesh
configurations, while the Utku element, although converging correctly, requires fine meshes
to converge completely.

The DKT element so complies with all its promises, particularly, its implementation on a
mini- or micro-computer is significantly easier than is the TRIC element. The TRIC element,
however, considers shear deformation, and not the DKT element, confining its use to thin
plates and shells.

33
Figure 9

34
Appendix 1. Derivation of the shape functions, {Hx} and {Hy}
In order to give it the deserved emphasis, we go back to the second assumption that we
alluded to previously. This assumption consists in imposing the Kirchhoff hypothesis at six
discrete points, which are the corner and the mid-nodes:

x = −dw/dx at the corner nodes 1, 2, 3

y = −dw/dy at the corner nodes 1, 2, 3 (a1.1)

sk = −(dw/ds)k at the mid-nodes defined anticlockwise around the element

boundary, figure a1, k = 4, 5, 6.

Figure a1

We thirdly assume that the transverse deflection, w, varies cubically along the three sides of
the triangle. The transverse displacement, w, along the triangle sides is thus approximated
by the same Hermite shape functions as those used for the Euler-Bernoulli beam element:

w(s) = N1.wi + N2.(−dw/ds)i + N3.wj + N4.(−dw/ds)j

where i, j are the nodes of the triangle side ij, and where the shape functions are the
Hermite shape functions:

N1 = 1 – 3(s/Lij)2 + 2(s/Lij)3 N2 = – s.(1 − s/Lij)2

N3 = 3(s/Lij)2 − 2(s/Lij)3 N4 = – s.[(s/Lij)2 – s/Lij]

Lij has already been defined as the length of the triangle side comprised between nodes i
and j, and s varies from zero at node i to Lij at node j.

The derivative of w with respect to s is thus given by

35
 dw/ds = (−6s/L2ij + 6s2/L3ij).wi + (1 − 4s/Lij + 3s2/L2ij).(dw/ds)I + (6s/L2ij − 6s2/L3ij).wj +
(−2s/Lij + 3s2/L2ij).(dw/ds)j

which takes the following expression at the mid-node, k, of the side ij

(dw/ds)k = −(3/2Lij).wi − (1/4).(dw/ds)I + (3/2Lij).wj − (1/4).(dw/ds)j (a1.2)

The last hypothesis to establish the KDKT stiffness matrix is that the rotations n, normal to
the triangle sides, vary linearly along the triangle side, that is,

(n)k = [(n)I + (n)j]/2 (a1.3)

Starting from the first assumption defined in (1) and (2), the other three assumptions enable
to express the two rotations, x and y, in function of the 9 pairs of shape functions of Hx and
Hy along with vector {ue}, which contains the 9 displacements at the corner nodes. For x, we
have so the following steps that summarize this derivation.

x = N1.x1 + N2.x2 + N3.x3 + N4.x4 + N5.x5 + N6.x6

or x = N1.y1 + N2.y2 + N3.y3 + N4.x4 + N5.x5 + N6.x6 (a1.4)

It is so required to express x4, x5 and x6 in function of the displacements at the two corner
nodes comprising nodes 4, 5, and 6, respectively, that is,

x4 = m41.w2 + m42.x2 + m43.y2 + m44.w3 + m45.x3 + m46.y3 (a1.5a)

x5 = m51.w3 + m52.x3 + m53.y3 + m54.w1 + m55.x1 + m56.y1 (a1.5b)

x6 = m61.w1 + m62.x1 + m63.y1 + m64.w2 + m65.x2 + m66.y2 (a1.5c)

so that {Hx} in x = HTx.{ue} would take the form previously given in (4):

m54.N5 + m61.N6
m55.N5 + m62.N6
N1 + m56.N5 + m63.N6
m41.N4 + m64.N6
Hx = m42.N4 + m65.N6 (a1.6)
N2 + m43.N4 + m66.N6
m44.N4 + m51.N5
m45.N4 + m52.N5
N3 + m46.N4 + m53.N5

In order to establish Hy, the same procedure will be applied for y. We will limit ourselves to
illustrate for x4 the derivation of these coefficients used in (a1.5) and (a1.6). x4 is expressed
in (a1.5a) in function of the displacements at the corner nodes 2 and 3 comprising mid-node
4. We make this derivation developing the geometric relation

36
 x4 = Cos23.n4 − Sin23.s4 (a1.7)

where 23 is the angle formed by the normal to the side 23 and the x-axis (figure a1). n4 will
be then developed using assumption (a1.3) and s4 assumptions (a1.2) and (a1.1).

a1.1. n4 development

By the fourth assumption (a1.3), we have

(n)4 = [(n)2 + (n)3]/2

and by the basic geometric relations

n2 = Cos23.y2 − Sin23.x2

n3 = Cos23.y3 − Sin23.x3

we get

n4 = 1/2(Cos23.y2 − Sin23.x2 + Cos23.y3 − Sin23.x3) (a1.8)

a1.2. s4 development

By the third assumption (a1.2), we have

(dw/ds)4 = −(3/2Lij).w2 − (1/4).(dw/ds)2 + (3/2Lij).w3 − (1/4).(dw/ds)3

or applying the second assumption (a1.1) to impose the Kirchhoff constraint at the nodes

(s)4 = (3/2Lij).w2 − (1/4).(s)2 − (3/2Lij).w3 − (1/4).(s)3

and by the basic geometric relations

s2 = − Cos23.x2 − Sin23.y2

s3 = − Cos23.x3 − Sin23.y3

we get

(s)4 = (3/2L23).w2 + (1/4).(Cos23.x2 + Sin23.y2) − (3/2L23).w3 + (1/4).(Cos23.x3 +

Sin23.y3) (a1.9)

We so introduce (a1.8) and (a1.9) in (a1.7) to get the expression for x4:

x4 = Cos23/2.(Cos23.y2 − Sin23.x2 + Cos23.y3 − Sin23.x3)

− Sin23/2.[(3/L23).w2 + (1/2).(Cos23.x2 + Sin23.y2) − (3/L23).w3 + (1/2).(Cos23.x3 +
Sin23.y3)] (a1.10)

and six of the eighteen coefficients of the second order shape functions, N1 to N6, in Hx (a1.6)
can be determined as the coefficients of w2, x2, y2, w3, x3, and y3 in (a1.10):

m41 = −(3/2L23).Sin23 m44 = (3/2L23).Sin23

m42 = m45 = −(3/4).Sin23.Cos23 m43 = m46 = (1/4).(2Cos223 − Sin223)

37
As we have Cosij = −yij/Lij Sinij = xij/Lij

These six coefficients become

m41 = −(3/2L223).x23 = 3a4/2 m44 = (3/2L223).x23 = −3a4/2

m42 = m45 = (3/4L223).x23.y23 = b4 m43 = m46 = (1/4L223).(2y223 − x223) = −c4

which comply with the Hx coefficients given in (4).

38
Appendix 2. Derivation of KeDKT from the bending strain energy
According to Batoz et al. (1980) the stiffness matrix of the DKT element for the analysis of
thin plates is based only on the expression of the bending strain energy:

∫
Ub= 1/2 {K}T[Db]{K}dxdy ==> ∫
KeDKT = [B(x, y)]T[Db][B(x,y)]dxdy

∫ ∫
==> KeDKT = 2A 0 to 1 0 to 1 - [B( )]T[Db][B( )]dd

where [B( )] is the matrical operator, which approximates {K} in function of the
displacements

{K} = [B( )].{ue}

The above expression for Ub can be demonstrated reversely developing the expression
below the integral of Ub. Omitting D16 and D26, we have

dx/dx dy/dy dx/dy + dy/dx D11 D12 dx/dx

D12 D22 dy/dy =
D66 dx/dy + dy/dx

D11.dx/dx + D12.dy/dy D12.dx/dx + D22.dy/dy D66.dx/dy + D66.dy/dx dx/dx

dy/dy
dx/dy + dy/dx

= D11.(dx/dx)2 + D12.dy/dy.dx/dx + D12.dx/dx.dy/dy + D22.(dy/dy)2 + D66.(dx/dy)2 +

D66.dx/dy.dy/dx + D66.dx/dy.dy/dx + D66.(dy/dx)2

==>
∫
Ub = 1/2 {D11.(dx/dx)2 + 2D12.dy/dy.dx/dx + D22 .(dy/dy)2 + D66.[(dx/dy)2 +
2.dx/dy.dy/dx + (dy/dx)2]}dxdy

and we recover the explicit expression of the bending strain energy for an orthotropic
material:

∫
Ub = 1/2 {D11.(dx/dx)2 + 2D12.dy/dy.dx/dx + D22 .(dy/dy)2 + D66.(dx/dy + dy/dx)2}dxdy

39
Appendix 3. The obscure side of the DKT element
The objective of this appendix is to shed some light on the obscure side of the DKT element,
which is the absence of an approximation for w within the element. We recall that to derive
the DKT bending stiffness matrix it was only required to develop the left-hand side of the
weak form. The complete weak form would be however

∫
− d2v/dx2.Mx.dA − ∫d v/dy .M .dA − 2∫d v/dxdy.M .dA = ∫v.f(x,y).dA
2 2
y
2
xy

+ ∫v.Q .dy
x + ∫v.Q .dx − ∫dv/dx.M .dy − ∫dv/dy.M .dx −
y x y

∫dv/dx.M xy.dx − ∫dv/dy.M .dy

xy (a3.1)

Although it was possible to replace, in the left-hand side of the weak form, the derivatives of
the weight functions with the 9 pairs of shape functions used to approximate x and y, and
so derive the KeDKT bending stiffness matrix, the absence of an approximation for w to
replace the weight function, v, in the right-hand side of the weak form, seeds at least two
doubts. The first question would be how rigorously could be formulated the load vector not
being possible to compute the integral

∫v.f(x,y).dA (a3.2)

Batoz et al. (1980) suggest bluntly in equation (8) that the load be shared equally by the
three corner nodes, and this has indeed given good results as far as relatively simple
loadings have been considered by the authors and ourself in our numerical examples, e.g.,
with uniform or punctual loads. In order to verify the behavior of the DKT bending element
with more complex types of loading, we have considered a sinusoidal load applied on a
isotropic plate, for which exists an analytical solution: in article 27 of their "Theory of Plates
and Shells", Timoshenko and Woinowsky-Krieger assume that the load distributed over a
rectangular plate is given by the expression

q = q0.Sin(x/A).Sin(y/B) (a3.3)

in which q0 represents the intensity of the load at the center of the plate, A and B are the
lengths of the plate sides, and the coordinates x and y vary from 0 to A and B, respectively.
Analytical expressions for the plate deflection, the bending moments, and the reactions on
the side of the plate are then easily derived in article 27. In this case of a sinusoidal load, we
may envisage different means to compute the load vector for the DKT element:

1. DKT (1): the pressure acting at the centroid of the element is computed by (a3.3),
multiplied by its area, and the result shared equally by the three nodes, similarly to
what is suggested by formula (8).
2. DKT (2): the pressure acting at each individual node is computed, multiplied by the
third of the element area, and the result introduced in the load vector at the first
position corresponding to this node; all other entries in the load vector being again
kept equal to zero.

40
 3. DKT (3): assume adhoc shape functions to replace v in (a3.2), for example, borrowed
from another 9-node triangular element. This route would require a deep
investigation, and is out of the scope of this paper.

The stresses and central displacement of the plate sinusoidally loaded were so computed at
the same locations considered anteriorly for the angle-ply laminate. A quarter of a square
isotropic plate, simply supported, was discretized from 4 x 4 to 20 x 20, and the two load
vectors, labeled DKT (1) and DKT (2), were tested. The convergences that we so obtained are
shown in figure a3.1.

Figure a3.1

Although sharing equally the load among the nodes converges slightly better, the important
observation is that the two cases converge from opposite sides; this suggests that may exist
a set of nine adhoc shape functions, which would optimize the convergence, justifying
researches in the sense of the DKT (3) route.

In the finite element method, the weak form is not just a nice artifice to reduce the order of
differentiation required in the shape functions, but is usually needed to make appear the
boundary conditions in the global matrix equation. In structural mechanics, the load shall be
in equilibrium with the reactions at the supports of the structure being studied. This
fundamental requirement shall appear in the global matrix equation and, thus, shall be
provisioned when the weak form is formulated. In order to achieve this requirement, all the
forces and moments acting on the sides of the element shall be produced in the weak terms
of the weighted residual statement. It is generally much more efficient to write the weak
forms from the differential equations that express the equilibrium of the forces and
moments acting on an infinitesimal portion of the structure. Doing so, the forces and
moments arising from these boundary terms, can be, with the load, isolated in the right-
hand side of the element matrix equation, while all integrals, which do not involve the load,
are kept in the left-hand side. Finally, the forces and moments, in the left-hand side integrals
of the equation only, are expressed in terms of the displacements. The forces and moments

41
remaining in the right-hand side of the element equation will be concentrated forces and
moments occurring at the nodes, and, upon global matrix assembly, will express the
reactions at the supports of the structure. Therefore, the second question, which arises from
the weak form (a3.1) applied to the DKT bending element, is how correctly this element
allows assessing the reactions on the side of the plate. The procedure to determine these
reactions is briefly remembered as follows.

We write the global matrix equation as

[KG]{uG} = {fG} + {QG} (a3.4)

where {QG} results from the assembly of all the integrals in the right-hand side of the weak
form, except integral (a3.2), which produces the global load vector (fG). Vector (QG}, usually
referred to as the global nodal forces vector, exhibits zero entries on the plate internal
nodes, and collects the reactions at the nodes situated on the boundary of the plate. After
having solved the global matrix equation, the global nodal load vector can be easily
established in a postprocessing operation making

{QG} = [KG]{uG} − {fG} (a3.5)

All nonzero entries in {QG} are forces and moments concentrated at the nodes situated on
the boundary of the plate. The reactions, expressed in forces and moments per unit length,
are so directly obtained dividing these entries by the length of the element aligned with the
boundary.

Timoshenko and Woinowsky-Krieger give us, in article 30 of their "Theory of Plates and
Shells", the reactions acting on the edges of a simply supported and uniformly loaded plate.
From the expressions given in article 30, it is verified that the sum of the distributed
reactions on the sides of the plate is larger than the total load applied on it. This apparent
incongruence is easily clarified with the consideration of the concentrated opposite reaction,
which surges at the corner of the plate. The distributed and the concentrated reactions that
act on the edges of the plate, and secure the load, are shown schematically in figure a3.2.

Figure a3.2

42
With DKT triangular and CPT(C) rectangular elements, we calculated by way of (a3.5) the
reactions on the quarter of a square plate discretized 20 x 20. The conforming four-node
CPT(C) element, with (w, dw/dy, dw/dx and d2w/dxdy) as nodal degrees of freedom,
requires a complete quartic polynomial, and, with the 20 x 20 discretization, produces
reactions virtually exact. The results of this exercise are presented in figure a3.3, whose
ordinates are the nondimensional reactions obtained dividing the actual reactions, for
example in N/m, by the product of the load with the side of the plate, that is, Vx = Vx/qA. The
value in the chart labeled DKT (*) at the middle of the side of the plate was calculated,
instead of from {QG}, resorting to the second derivatives of x and y, which for the DKT
element are constant over the element:

Vx = Qx + dMxy/dy (a3.6)

where for an orthotropic material, we have

Qx = D11.d2x/dx2 + (D12 + D66).d2y/dxdy + D66.d2x/dy2 (a3.6.1)

dMxy/dy = D66(d2x/dy2 + d2y/dxdy) (a3.6.2)

so that Vx = D11.d2x/dx2 + (D12 + 2D66).d2y/dxdy + 2D66.d2x/dy2 (a3.6.3)

Figure a3.3

The reaction at the corners of the plate is red directly in {QG}, for example in newton; and, its
nondimensional expression is calculated by R = R/qA2, whose values obtained with the two
elements are compared below with the analytical solution.

R
Analytical 0,0650
DKT 0,0640
CPT(C) 0,0642

43
Although the conjunction of the distributed and concentrated reactions calculated from (QG}
keeps the load exactly, the DKT element produces an important divergence in the reaction
at the middle of the sides of the plate, which is expected to be a maximum, but is obtained
some 30% lower than the analytical solution (*). A much better estimate for that reaction
can be however computed with formulas (a3.6). The discrepancy in vector {QG} proceeds
directly from this obscure side of the DKT element of not having an approximation for w
defined. An accurate assessment of the reactions is, however, required to design how the
structure should be supported by means of whatever material. The engineer should be thus
aware of this defect, and must be attentive to the possible occurrence of inaccurate values
in vector {QG} so that he can decide to make complementary calculations with formulas
(a3.6).

(*): It may be argued that a full plate discretization would eliminate this defect of inaccurate
reactions at the symmetry lines of the plate. This possible solution of the problem would,
however, bear a heavy price in terms of computational performances.

44